How OpenAI stress-tests its large language models
Description
OpenAI is once again lifting the lid (just a crack) on its safety-testing processes. Last month the company shared the results of an investigation that looked at how often ChatGPT produced a harmful gender or racial stereotype based on a user’s name